Lecture 20
[[lecture-data]]
2024-10-14
Readings
- a
7. Chapter 7
Recall from singular value decomposition we have that for any matrix
for some
The values in
For any
Further, if
(we take "diagonal" to mean that the only entries that can be nonzero have the same row and column index)
- We can think of this as a generalization of diagonalization/spectral decomposition, but where the loss is that
and are distinct. - if
is positive semidefinite, then the diagonalization is the singular value decomposition since we have that for some unitary and nonnegative all real.
If
If
For all
When
Let
(see polar decomposition)
Say
- easy to see from (1) - easy to see from (1) - For any
, we have
To see (5) and (6) , recall that
To see (7), note that
- Note
is by Rayleigh-Ritz that this equals the largest eigenvalue of . - Then
is by the fact that the eigenvalues of - ie the eigenvalues are the squared singular values of .
recall that
Let
For each
- if
satisfies 1, 2, then is a 1-2-generalized inverse
If
- preview: every matrix has a unique 1-2-3-4-generalized inverse. We call this the Moore-Penrose inverse
If
(see generalized inverse)
Consider
But if
Let
Say
(see 1-generalized inverses give solutions to consistent linear systems)